Anthropogenic fugitive, combustion and industrial dust is a significant, underrepresented fine particulate matter source in global atmospheric models

Global measurements of the elemental composition of fine particulate matter across several urban locations by the Surface Particulate Matter Network reveal an enhanced fraction of anthropogenic dust compared to natural dust sources, especially over Asia. We develop a global simulation of anthropogenic fugitive, combustion, and industrial dust which, to our knowledge, is partially missing or strongly underrepresented in global models. We estimate 2–16 μg m−3 increase in fine particulate mass concentration across East and South Asia by including anthropogenic fugitive, combustion, and industrial dust emissions. A simulation including anthropogenic fugitive, combustion, and industrial dust emissions increases the correlation from 0.06 to 0.66 of simulated fine dust in comparison with Surface Particulate Matter Network measurements at 13 globally dispersed locations, and reduces the low bias by 10% in total fine particulate mass in comparison with global in situ observations. Global population-weighted PM2.5 increases by 2.9 μg m−3 (10%). Our assessment ascertains the urgent need of including this underrepresented fine anthropogenic dust source into global bottom-up emission inventories and global models.


Introduction
Outdoor PM 2.5 (fine particulate matter with aerodynamic diameter less than 2.5 micrometers) is the fifth largest risk factor for premature mortality worldwide (Forouzanfar et al 2016). Global atmospheric models are widely used for assessments of exposure to outdoor dust. The latter includes three broad categories, mineral dust naturally windblown from arid desert regions (Prospero et al 2002), anthropogenic windblown dust from human disturbed soils due to changes in land use practices, deforestation and agriculture (Tegen et al 1996(Tegen et al , 2004, and anthropogenic fugitive, combustion, and industrial dust (AFCID) from urban sources. Global models typically include natural mineral dust (Huneeus et al 2011, Astitha et al 2012 with recent developments to assess the relative contribution of anthropogenic windblown Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
dust (Ginoux et al 2012, Huang et al 2015, Guan et al 2016. However, to our knowledge, AFCID is partially missing or strongly underrepresented from global models (Rind et al 2009) as evident from model descriptions published as part of several multi-model inter-comparison studies (Schulz et al 2006, Myhre et al 2013, Pan et al 2015, Silva et al 2013, Huneeus et al 2011. Measurements of PM 2.5 and its chemical composition over several urban locations by the Surface Particulate Matter Network (SPARTAN) offer information about PM 2.5 sources (Snider et al 2015(Snider et al , 2016. Snider et al (2016) found an enhanced fraction of AFCID compared to natural sources over several Asian cities, evidenced by a high zinc (mainly anthropogenic as evidenced by Councell et al 2004 andHarrison et al 2012) to aluminum (mainly natural) ratio in PM 2.5 dust. Sources of anthropogenic fugitive, combustion, and industrial dust include elemental components from coal combustion (fly ash) and industrial processes (e.g. iron and steel production, cement production), resuspension from paved and unpaved roads, mining, quarrying, and agricultural operations, and road-residential-commercial construction (McElroy et al 1982, Watson and Chow 2000, Guttikunda et al 2014. Some evidence for the significance of these anthropogenic fugitive, combustion, and industrial sources to ambient PM 2.5 dust is emerging through measurements and source apportionment studies (Yang et al 2011, Yu et al 2013, Viana et al 2008, Mooibroek et al 2011. Despite the majority of these emissions being in the coarse mode there is a tail that contributes to PM 2.5 . AFCID includes several trace elements that are associated with adverse health effects, but not yet well understood (West et al 2016).
The few global emission inventories that include anthropogenic primary emissions of total PM 2.5 have limited distinction between estimates of fugitive, combustion and industrial dust, and rather incomplete representation of fugitive sources (e.g. Janssens However, the contribution of AFCID sources to PM 2.5 mass remains poorly quantified, especially at the global scale.
Several global and regional models tend to consistently underestimate aerosol loading (Moorthy et al 2013, Pan et al 2015, Lelieveld et al 2015, Brauer et al 2016. We hypothesize that inclusion of missing AFCID sources will reconcile some of the unexplained bias. Here, we develop a global simulation of anthropogenic fugitive, combustion, and industrial dust, and evaluate it with in situ measurements.

Materials and methods
We interpret Surface Particulate Matter Network (www. spartan-network.org) measurements of PM 2.5 and trace metals collected from monitoring stations over geographically diverse global regions to evaluate our simulation of AFCID (Snider et al 2015(Snider et al , 2016. SPARTAN measurements include an AirPhoton SS4i automated air sampler to collect aerosol on PTFE filters for gravimetric assessment of PM 2.5 mass, and Inductively Coupled Plasma-Mass Spectrometry to quantify PM 2.5 trace metals used to determine crustal PM 2.5 (Snider et al 2016). Measurement sites are primarily in urban locations with site selection designed for spatial representativeness. SPARTAN measurements exhibit a high degree of consistency with independent measurements over Asia (Beijing, Bandung, Kanpur and Hanoi), the U.S. (Mammoth Cave and Atlanta) and elsewhere (Snider et al 2015(Snider et al , 2016. We obtain global monthly mean anthropogenic emissions of primary particulate matter (including fugitive, combustion, and industrial dust) in 2015 from the ECLIPSE dataset (version V5a; www.iiasa.ac.at/web/ home/research/researchPrograms/air/Global_emis sions.html). Klimont et al (2016)  The anthropogenic primary PM 2.5 emission inventories are derived using a dynamic technologybased approach employing high source-activity-sector resolution at a country or even subnational level. For each of the emission sources, the models applied to calculate these inventories define activity rate, unabated emission factors, penetration and removal efficiency of applicable emission control technologies (Lei et al 2011, Klimont et al 2016). The data and assumptions used in the inventories draw on international and national statistics, on an array of measurement studies representative for typical sources and applied technologies considering local circumstances and studies, and on information about the air quality legislation and efficiency of its enforcement allowing defining of the penetration of control measures. These inventories include a harmonized calculation of mass-based size distribution (PM 2.5 , PM 10 ) and primary carbonaceous aerosols. The characteristics of sources vary strongly with respect to the contribution of carbonaceous particles and the underlying models capture these features by defining mass-based consistent emission factors and removal efficiencies for total PM 2.5 , black carbon, organic carbon and particulate organic mass. Compared to previous global work, ECLIPSE includes estimates for a number previously unaccounted or often underestimated PM sources, that is, gas flaring, kerosene lamps, diesel generators (Klimont et al 2016).
We We use the HEMCO module (Keller et al 2014) to implement the AFCID emission inventory into GEOS-Chem. We conduct simulations from January 1, 2014 to December 31, 2015 following a one month spin-up. We use operator durations of 10 min for transport and 20 min for chemistry for optimized computational speed and accuracy (Philip et al 2016). We calculate ground-level PM 2.5 at 35% relative humidity to follow common measurement protocols. We convert organic carbon to particulate organic mass following Philip

Results and discussion
The top panel of figure 1 shows filled concentric circles of campaign-mean PM 2.5 dust (inner circles) measured by the SPARTAN network over 13 globally dispersed locations, for the years 2013-2015 (Snider et al 2016). SPARTAN dust mass (and % of total PM 2.5 ) varies from ∼1 mg m À3 (∼10%) over North America, ∼5 mg m À3 (5%-15%) over South and South East Asian cities (Kanpur, Dhaka, Hanoi) to ∼14 mg m À3 (∼25%) over Beijing (Snider et al 2015(Snider et al , 2016. Enhanced Zn:Al ratios measured over these sites provide evidence of an anthropogenic source (Snider et al 2016). The middle panel of figure 1 shows the GEOS-Chem simulated natural mineral dust. Natural mineral dust concentrations are enhanced over regions with accumulated alluvial sediments, predominantly over arid and semiarid regions of North Africa, the Middle East and Central Asia (Zender et al 2003, Fairlie et al 2007, Huneeus et al 2011. It is evident that the pronounced dust concentrations measured over East and South Asia cannot be explained by natural mineral dust alone (Lei et al 2011. The bottom panel of figure 1 shows the simulation of anthropogenic fugitive, combustion, and industrial dust. AFCID increases PM 2.5 dust concentrations by 2-16 mg m À3 over much of East and South Asia. The concentration of simulated AFCID is comparable to that of natural mineral dust over parts of Europe and Eastern North America. Other regional studies (Appel et al 2013, Park et al 2010) offer additional evidence of AFCID sources.
The top panel of figure 1 shows that GEOS-Chem simulated AFCID in addition to default natural mineral dust reduces the bias in total dust mass measured at SPARTAN sites over Asia. A high AFCID over Beijing reveals the significance of regional fugitive sources (Yu et al 2013. Zhang et al (2015) use the adjoint of GEOS-Chem together with the MEIC inventory to attribute 27% of wintertime PM 2.5 over Beijing from emissions of AFCID from North China. Table 1 contains statistics describing the comparison of GEOS-Chem simulated concentrations versus in situ observations. The inclusion of AFCID increases the correlation versus PM 2.5 dust mass concentration from 0.06 to 0.66 over all SPARTAN sites compared to Environ. Res. Lett. 12 (2017) 044018 campaign-mean data. A test case study that excludes two arid sites (Ilorin, Nigeria and Rehovot, Israel) dominated by large simulated natural mineral dust loading also reveals an improved consistency from slope ¼ 0.29 (r ¼ 0.77) to slope ¼ 1.29 (r ¼ 0.91) further demonstrating the importance of AFCID at the global scale. Figure 2 shows the in situ and simulated concentration of total PM 2.5 . The top panel shows enhanced PM 2.5 concentrations in the in situ measurements over rapidly developing Asia. The bottom panel shows that the simulation with AFCID largely reproduces these enhancements. We find that simulated AFCID comprises 5%-15%  of total PM 2.5 across large parts of East and South Asia. Table 1 quantifies the comparison of GEOS-Chem simulated PM 2.5 concentrations versus long-term annual mean in situ measurements compiled by Brauer et al (2016) for the Global Burden of Disease Study. Site locations span a diversity of environments including routine monitoring networks in both densely populated and remote areas. The additional PM 2.5 source from AFCID increases the slope of the best fit line from 0.83 to 0.93. This analysis reveals that neglect of AFCID in PM 2.5 can underestimate ambient PM 2.5 concentrations by 5%-10% globally, and by up to 15% in East and South Asia. Global populationweighted PM 2.5 concentrations increase by 2.9 mg m À3 (10%) with implications for future assessments of PM 2.5 health effects.

Conclusions
PM 2.5 health impact assessments require a complete description of PM 2.5 sources. We interpret global crustal PM 2.5 observations from the SPARTAN network and find evidence for anthropogenic fugitive, combustion, and industrial dust. A collection of emission inventories (ECLIPSE, IIT-B and MEIC) was used to estimate AFCID emissions for inclusion into a GEOS-Chem simulation. Inclusion of AFCID increased total PM 2.5 mass by 2-16 mg m À3 over anthropogenic polluted regions across East and South Asia, reducing the observed bias from 17% to 7% in comparison with the global PM 2.5 in situ observations, and increasing the correlation from 0.06 to 0.66 of PM 2.5 dust concentration compared to SPARTAN in situ observations. Global population-weighted PM 2.5 concentrations increase by 2.9 mg m À3 (10%). The noteworthy contribution of this underrepresented AFCID source to PM 2.5 mass as evaluated with observations, motivate further development and incorporation of AFCID emission into global models. To our knowledge, this is the first global assessment of the importance of anthropogenic fugitive, combustion, and industrial dust. Nonetheless some portion of this anthropogenic dust source might not be captured well in our inventories, with potential uncertainty in our estimates. Future work should assess the implications of coarse mode AFCID that may be associated with the PM 2.5 examined here. Although we focus on the ground-level PM 2.5 owing to its importance in human health impact studies, estimating AFCID and understanding its optical and transport properties could benefit studies of climate forcing (Rind et al 2009) and visibility.

Acknowledgments
We are thankful to Christoph Keller, Brian Boys, Melanie Hammer, Chi Li and Pankaj Sadavarte for helpful discussions that improved the manuscript. This work was supported by the Natural Sciences and Engineering Research Council of Canada. The SPARTAN network is an International Global Environ. Res. Lett. 12 (2017) 044018